[AMD][AgentX] Port MI355X DeepSeek V4 MTP disagg to srt-slurm - #2555

Open

cquil11 wants to merge 23 commits into

mainfrom

agent/port-dsv4-mi355x-agentx-srt-slurm

Commits on Aug 31, 2026

refactor(amd): centralize MI355X srt-slurm runtime
cquil11
committed
0a89453
fix(amd): stream srt-slurm stdout in Actions
cquil11
committed
0674eaa
fix(amd): preserve fixed-seq launch semantics
cquil11
committed
6439c63
fix(amd): reclaim stale hugepages before srt workers
cquil11
committed
d0b0f31
fix(ci): reclaim legacy AMD runner files before checkout
cquil11
committed
429de0f
fix(amd): pin lm-eval pip compatibility
cquil11
committed
190a387
fix(amd): pin eval framework runtime
cquil11
committed
74c0c85
fix(amd): preserve legacy AgentX fabric semantics
cquil11
committed
c94d0a4
fix(amd): pin writable lm-eval cache runtime
cquil11
committed
2a10192
fix(amd): retain srt lm-eval result artifacts
cquil11
committed
847c80f
fix(amd): use srt lm-eval result contract
cquil11
committed
539f2fa
Merge remote-tracking branch 'origin/main' into codex/base-pin-eval
cquil11
committed
09f2299
feat(amd): port DSV4 AgentX to srt-slurm
cquil11
committed
0bc3851
fix(mi355x): bound DSV4 TP8 HiCache host memory
cquil11
committed
4e5108f
fix(mi355x): preserve proven DSV4 serving config
cquil11
committed
3d556e0
fix(amd): preserve AgentX eight-hour job budget
cquil11
committed
77480bd
fix(amd): preserve DSV4 AgentX runtime contract
cquil11
committed
1b71cb4
fix(amd): record writable eval runtime pin
cquil11
committed
e1329a4
fix(amd): stage srt eval result artifacts
cquil11
committed
aed36b6